The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted student input and stores it in markdown files (XX_标题.md, 学生画像.md) which are subsequently read back by the agent to adapt teaching strategies and update the student profile. This creates an indirect prompt injection surface where a student could provide answers containing instructions aimed at influencing the agent's future behavior.
Ingestion points: Student answers are recorded in course files under the ## 学生回答 header and extracted into the 学生画像.md file.
Boundary markers: The skill uses markdown headers as structural delimiters but lacks explicit instructions for the agent to ignore or sanitize embedded commands within the student answer sections when re-reading them.
Capability inventory: The skill possesses file system read/write and directory creation capabilities to manage its learning environment.
Sanitization: No specific sanitization or escaping of student input is mentioned before it is committed to long-term storage or used in profiling logic.

interactive-learning